The Roma/non-Roma Test Score Gap in Hungary

Transcription

1 BUDAPEST WORKING PAPERS ON THE LABOUR MARKET BWP 2010/10 The Roma/non-Roma Test Score Gap in Hungary GÁBOR KERTESI - GÁBOR KÉZDI INSTITUTE OF ECONOMICS, HUNGARIAN ACADEMY OF SCIENCES DEPARTMENT OF HUMAN RESOURCES, CORVINUS UNIVERSITY OF BUDAPEST BUDAPEST, 2010

2 Budapest Working Papers On The Labour Market BWP 2010/10 Institute of Economics, Hungarian Academy of Sciences Department of Human Resources, Corvinus University of Budapest The Roma/non-Roma Test Score Gap in Hungary Authors: Gábor Kertesi senior research fellow Institute of Economics of the Hungarian Academy of Sciences Gábor Kézdi associate professor Central European University senior research fellow Institute of Economics of the Hungarian Academy of Sciences December 2010 ISBN ISSN

3 The Roma/non-Roma Test Score Gap in Hungary Gábor Kertesi - Gábor Kézdi Abstract This paper documents and decomposes the test score gap between Roma and non-roma 8th graders in Hungary in Our data connect national standardized test scores to an individual panel survey with detailed data on ethnicity and family background. The test score gap is approximately one standard deviation for both reading and mathematics, which is similar to the gap between African-American and White students of the same age group in the U.S. in the 1980s. After accounting for on health, parenting, school fixed effects and family background, the gap disappears in reading and drops to 0.15 standard deviation in mathematics. Health, parenting and schools explain most of the gap, but ethnic differences in those are almost entirely accounted for by differences in parental education and income. JEL: I20, J15 Keywords: test score gap, Roma minority, Hungary Acknowledgement: Funding from OTKA-68523K project is gratefully acknowledged. 3

4 Roma és nem roma tanulók teszteredményeinek különbsége Kertesi Gábor - Kézdi Gábor Összefoglaló A tanulmány bemutatja a magyarországi roma és nem roma nyolcadikosok 2006-ban mért teszeteredményeinek átlagos különbségét, és felbontja azt egyéb változóknak betudható és pusztán etnikai különbségekre. Az etnikai összehasolítást az Országos Kompetenciamérés és az Életpálya felmérés adatainak az összekapcsolása teszi lehetővé. A roma és nem roma nyolcadikosok között mért átlagos különbség egy szórásegység körüli mind a matematika mind a szövegértés teszteredményekben. Ez a különbség nagyon hasonló ahhoz, amit hasonló korú fekete és fehér tanulók között mértek az Amerikai Egyesült Államokban a 80-as évek elején. Az egészségi állapotra, az otthoni nevelési környezet változóira, iskola fix hatásokra, valamint a szülők iskolázottságára és jövedelmi viszonyaira kontrollálva a roma és nem roma tanulók közötti különbségek a szövegértés teszteredményben teljes mértékben eltűnnek, és a matematika teszteredményben is nagymértékben, 0.15 szórásegységre csökkennek. Az egészség, az otthoni nevelési környezet és az iskola a teszteredmények etnikai különbségeinek nagy részét megmagyarázzák. Az egészségi körülményekben és az otthoni nevelési környezetben meglevő, jelentős mértékű etnikai különbségek ugyanakkor gyakorlatilag teljes mértékben betudhatók a szülők iskolai végzettségében és a család jövedelmi viszonyaiban meglévő különbségeknek. Tárgyszavak: teszteredmény különbségek, roma kisebbség JEL: I20, J15 Köszönetnyilvánítás: Köszönettel tartozunk az OTKA-68523K projektnek kutatásunk támogatásáért. 4

5 This paper documents and decomposes the test score gap between Roma and non-roma 8 th graders in Hungary in Our data connect national standardized test scores to an individual panel survey with detailed data on ethnicity and family background. The test score gap is approximately one standard deviation for both reading and mathematics, which is similar to the gap between African-American and White students of the same age group in the U.S. in the 1980s. After accounting for on health, parenting, school fixed effects and family background, the gap disappears in reading and drops to 0.15 standard deviation in mathematics. The Black-White test score gap has been a subject of intensive research in the United States. The Educational Testing Service (2010) provides a comprehensive overview of the time series of the test score gap, and several studies analyze its causes and consequences (see, for example, Roland G. Fryer and Steven D. Levitt, 2006 and the volume edited by Katherine Magnuson and Jane Waldfogel, 2008). This literature finds that the gap increases across grades; in all grades it narrowed considerably until the 1980s, but after that time, the trend stopped or slowed. The residual gap in regressions with family background and parenting variables is zero or small in lower grades but remains substantial in upper grades. Our results allow a direct comparison to many of the findings of the Black-White test score gap literature. The Roma (also known as the Romani people or Gypsies) constitute one of the largest and poorest ethnic minority groups in Europe and are concentrated in the countries of Central and Eastern Europe. The size of the Roma population was about 4 million in the early 1990s (Zoltan Barany, 2002). Due to a high birth rate, the Roma population continues to grow, resulting in increasing population shares. In Hungary, the Roma are estimated to comprise 5 to 6 percent of the total population and 10 to 12 percent of the young adolescent population (István Kemény and Béla Janky, 2006). The Roma have resided in Central and Eastern Europe for centuries, but their history has been characterized by separation and exclusion. 5

6 Table 1. Selected social indicators for the Roma and the non-roma in Hungary, and African-Americans and Whites in the United States. Hungary United States Roma non- Roma Black White Education - secondary or more (percent of all adults) a,b,c Education - college or more (percent of all adults) a,b,c Employment to population ratio, men (percent of all adults) a,b,d Employment to population ratio, women (percent of all adults) a,b,d Unemployment rate (percent) d,e Live in rural area (percent) e,f Number of children born to women, age 15 to 19 a,g,f Number of children born to women, age 40 to 44 a,g,f Infants born with low birth weight (percent) e,h Percentage of children in single-parent families e,i a The Roma figures are estimates from the Hungarian Roma Survey of 2003 (Kemény and Janky, 2006). Age groups: 25 years and over for the education figures, 15 years and older and not in school for the employment figures. b The non-roma figures are overall national estimates from the Hungarian Labor Force Survey of Age groups: 25 years and over for the education figures, 15 years and older and not in school for the employment figures. c The U.S. figures are from published tables on the U.S. Census website ( Table 224. Educational Attainment by Race, and Hispanic Origin ), and they refer to Age group: 25 years and over. d The U.S. figures are from published tables on the BLS website ( Labor Force Statistics from the Current Population Survey ), and they refer to the fourth quarter in Age group: 16 years and over. e The Roma and non-roma figures are estimates from the Hungarian Life Course Survey (Kertesi and Kézdi, forthcoming), and they refer to eighth graders or the parents of eighth graders in f The U.S. figures are from published tables on the U.S. Census website ( Profiles of General Demographic Characteristics ), and they refer to g The non-roma figures are overall national figures from the published tables of the Hungarian Census of 2001 (Volume 22, table 1.3). h The U.S. figures are from Table 33 in the National Vital Statistics Reports, 58(24) (U.S. Census Bureau), and they refer to i The U.S. figures are from Table C9 in America s Families and Living Arrangements: 2009 (U.S. Census Bureau), and they refer to all children under 18 in Table 1 shows a comparison with some corresponding African-American figures from the United States. In terms of education and employment, the gap between Roma and non- Roma is substantially larger than the gap between African-Americans and Whites in the U.S. The Roma are somewhat more rural, and they have a substantially higher birth rate relative to the majority. The same is not true for African-Americans. The teen birth rate is higher and low birth weight is significantly more common among the Roma than the mainstream population, and the gaps are similar in magnitude to the Black-White gap. Single-parent 6

7 families are less frequent among the Roma in Hungary than among the majority, while they are substantially more frequent among African-Americans than among Whites in the U.S. I. DATA We use the test scores of 8 th -grade students measured by the Hungarian National Assessment of Basic Competences (NABC) in May 2006, which is linked to the sample of the Hungarian Life Course Survey (HLCS). The NABC measures the mathematical and reading literacy skills of entire cohorts of 6 th -, 8 th - and 10 th -grade students. The NABC does not cover students with special education needs 1 except for 8 th graders in The Hungarian Life Course Survey (HLCS), conducted by TARKI Research Institute, is an individual panel survey administered yearly that follows the model of the National Longitudinal Surveys of Youth in the United States (NLSY79). The original sample is 10,000 students drawn from the population of 8 th -grade students with valid test scores in May The sample includes students with special education needs (and their scores in reading). Results excluding students with special education needs are similar and presented in the online appendix. Students with lower test scores are overrepresented in the survey, and we use sampling weights to restore population moments. Our sample consists of students who were interviewed in the first two survey waves and who lived with at least one biological parent. These sample restrictions are necessary to identify ethnicity. Each of the first two waves includes two questions on ethnic or national identity. These question-pairs allowed parents to declare multiple identities, and many did so. In this paper, we consider as Roma all students whose (biological) mother or (biological) father chose Romani identity as a first or second choice in either of the two waves. According to this definition, the fraction of Roma students is close to 8 percent, and the size of the Roma subsample is The size of the sample is 9056 students for the reading test and 8335 for mathematics. This difference in sample size exists because students with special education needs have test scores in reading but not mathematics. The online appendix shows the number of observations lost due to the sample selection together with some descriptive statistics on the lost individuals. 1 Six percent of all 8 th graders (twelve percent of the Roma 8 th graders) in 2006 were students with special education needs; most of them were mildly mentally disabled. Most special education needs students do not participate in the NABC. In 2006, a special version of the reading test was administered to these students as well, and our data include those test scores. 2 The survey probably captures four fifths of the students who are considered Roma by their teachers. School principals estimated the fraction of Roma students in the entire primary school population (grades 1 through 8) to be 12 percent (NABC data), which translates to around 10 percent in 8 th grade. Alternative definitions of Roma ethnicity (both mather and father Roma, Roma is indicated in both survey waves, and similar combinations) give very similar results in all regressions. 7

8 II. THE TEST SCORE GAP Table 2 shows the standardized test score gap between Roma and non-roma 8 th graders in Hungary in 2006 as well as the gap between African-American and White students in the U.S. for a few selected years. The U.S. series are presented in two different groups because the published time series of 8 th graders begin in 1992 while the series for 13-year-olds begins in the late 1970s. The ethnic gap in Hungary is very similar to the Black-White gap among 13-year-old students in 1978/80. In both cases, the gap in reading is less than one standard deviation, while the gap in mathematics is greater than one standard deviation. Table 2. The Roma/non-Roma and Black-White test score gaps in Hungary and the U.S., respectively, among eight graders or 13-year-old students. Test scores are standardized by national standard deviations. Roma/non-Roma gap, 8 th grade, Hungary a Black-White gap, 8 th grade, U.S. b Black-White gap, age 13, U.S. c Reading Mathematics Reading Mathematics Reading Mathematics 1978/ / a The authors calculations using the National Assessment of Basic Competences of Hungary linked to the Hungarian Life Course Survey. b. National Assessment of Educational Progress (NAEP), Main NAEP tables, 1992 and c. National Assessment of Educational Progress (NAEP), Long-Term Trend tables, 1980, 1992 and 2008 in reading, 1978, 1992 and 2008 in mathematics. III. METHODOLOGY AND RIGHT-HAND SIDE VARIABLES We estimate a series of OLS regressions with the Roma dummy and control variables on the right-hand side. We start without controls and successively add measures of children s health, the parenting they experienced, school and class fixed effects and variables for family structure, parental education and permanent income. The main question is the extent to which the coefficient on the Roma dummy decreases with the inclusion of the control variables. Although all of our models are reduced-form regressions, the content of the control variables and the sequence of their inclusion suggest causal mechanisms that are in line with those found in previous literature. The ethnic gap in test scores may be caused by ethnic differences in health, parenting and schools, which represent the most important causal mechanisms through which differences in parental education and income may lead to large differences in test scores. 8

9 The first measure of health is a dummy for low birth weight (less than 2500 grams) as an indicator of fetal health status. Adverse fetal health status is shown to have substantive negative consequences for cognitive development in both the short run and the long run and is also highly correlated with poverty (Nancy Reichman, 2005; Jere R. Behrman and Mark Rosenzweig, 2004; Sandra Black, Paul J. Devereux and Kjell G. Salvanes, 2007). The second health measure is teenage body height in units of gender-specific standard deviations (with age correction). Body height is a standard marker of prenatal and childhood nutritional and health history (Anne Case and Christina Paxson, 2008). The third measure is a dummy for fair or poor subjective health status as reported in the first survey wave (at modal age 15). Evidence presented by Anne Case, Darren Lubotsky and Christina Paxson (2002) shows that reported health status correlates strongly with children s chronic conditions as assessed by physicians. Differences in parenting are likely to be important causal mechanisms underlying the ethnic test score gap. In their extensive review, Jeanne Brooks-Gunn and Lisa Markman (2005) conclude that parenting differences, particularly differences in language use, daily storybook reading and a cognitively stimulating home environment, play a crucial role. We have two sets of variables for parenting. The first set measures parenting practices in early childhood. These variables are based on retrospective questions that the parents and children were separately asked. Parents were asked about the frequency of activities that they engaged in with the child during the preschool years, for which we include dummies for the frequency of bedtime storytelling, visits to the theater and hiking. The child was also asked about the frequency of bedtime stories in a separate interview, and we enter two dummies for their frequency. The second set of parenting variables contains two standardized measures from the HOME inventory scale at modal age 15, the cognitive stimulation subscale and the emotional support subscale. Extensive research (Robert H. Bradley and al., 2000; Frank L. Mott, 2004) has demonstrated that HOME measures are highly correlated with cognitive and non-cognitive development and have predictive power for outcomes later in life. Our measures are derived from the Short Form (27 items) of the Early Adolescent version of the Home Observation for Measurement of the Environment (HOME-SF) for children aged years as applied in the NLSY. School quality is controlled for through the inclusion of school fixed effects. In another specification, class fixed effects are included (interacted with school fixed) to control for differences in exposure to teachers and peers. School choice is free in Hungary, which likely results in strong sorting by income and ethnicity. As a result, the schools and classes of Roma students may differ considerably from the schools and classes of non-roma students. School quality and teacher effectiveness are notoriously difficult to measure by observable characteristics. By entering fixed effects, we compare Roma and non-roma students within 9

10 the same schools and classes and can thus capture both the otherwise-measured and unmeasured differences in their experiences. The administrative source of the test score data includes identifiers for schools and classes, and the two-stage sampling procedure of the matched HLCS sample ensures that we have enough students in the sample who shared the same school and class in 8 th grade for a fixed effects analysis. At the same time, however, the majority of non-roma students do not share a school with Roma students in our sample. The last set of variables that we enter covers family structure, parental education and measures for permanent income that we consider pre-determined with respect to children s health, parenting environment and schools. The family structure variables include whether, at the time of the first interview (at modal age 15), students lived with their biological mother, biological father, stepmother or stepfather. In addition to variables for the mother s and the father s level of education, we include the number of books at home (in categories) and access to Internet at home. Permanent income measures are parents employment status, the fraction of years that they had been employed since the birth of the student, log household income, log household size, number of non-employed adults, size of the apartment both in terms of square meter per capita and number of rooms per capita, bathroom access, and five indicators of poverty (whether, in a 12-month period, the household felt that it had no money for food or heating, the household received welfare or the student received free schoolbooks and free lunches at school). We estimate seven specifications. After reproducing the raw gap without control variables, we first include the health measures, then measures of the home environment and then school and class fixed effects. Last, we add the family background variables, first without the school and class fixed effects, and then together with those effects. IV. REGRESSION RESULTS The Roma versus non-roma test score gap estimates from the seven specifications are presented in Table 3. The standard error estimates are robust to heteroskedasticity and clustering at the school level. Missing right-hand side variables are addressed by including dummies for missing status. The detailed results are in the online appendix.3 3 These are linear regressions and may suffer from functional form misspecification and lack of common support between the Roma and non-roma subsamples. We re-estimated specifications (2), (3) and (6) by nearest neighbor matching for the propensity score and got very similar results (see the online appendix). 10

11 The results are qualitatively similar across the two tests. Inclusion of health decreases the gap by 10 percent, and inclusion of home environment and parenting leads to a substantial further decrease of more than 50 percent in the case of reading and slightly less than 50 percent in the case of mathematics. Inclusion of school fixed effects decreases the gap by an additional third, and class fixed effects lead to a smaller but non-negligible further decrease. The combined reduction of the Roma dummy is large after the inclusion of these variables, which are intended to measure causal mechanisms. The ethnic gaps in reading and mathematics decrease to 0.16 and 0.28 of their standard deviations, respectively, indicating that ethnic differences in childhood health, home environment and schools can account for at least 75 to 85 percent of the ethnic gap in test scores in eighth grade. Addition of the rest of the family background variables but not the school and class fixed effects reduces the ethnic gap to 11 percent in reading and 22 percent in mathematics. After inclusion of all right-hand side variables, the gap becomes 5 percent (insignificant) in reading and 15 percent in mathematics. Table 3. The ethnic gap in reading and mathematics: unconditional and conditional on control variables. OLS estimates of the Roma coefficient in seven specifications. (1) (2) (3) (4) (5) (6) (7) Panel A. Reading Gap [S.E.] [0.05] [0.05] [0.05] [0.06] [0.07]* [0.05]* [0.07] Observations R Panel B. Mathematics Gap [S.E.] [0.05] [0.05] [0.05] [0.05] [0.07] [0.05] [0.07]* Observations R Control variables Health Yes Yes Yes Yes Yes Yes Home environment Yes Yes Yes Yes Yes School FE Yes Yes Yes School Class FE Yes Yes Family background Yes Yes 11

12 V. ETHNIC GAP IN HEALTH AND PARENTING Taking one step back, we also look at the ethnic gap in the most important measures of health and parenting. For each health and parenting variable, we estimate the raw gap (with the Roma dummy as the only variable on the right-hand side) and the conditional gap, which is the coefficient on the Roma dummy after inclusion of the family background variables (family structure, parental education and permanent income). The goal of this analysis is to determine ethnic differences in the most important variables that can have causal effects. A similar analysis for school and class fixed effects would be less straightforward. Table 4. Ethnic gap in health and parenting. Raw differences and differences conditional on family background variables. OLS results. Low birth weight Standardized height Fair or poor health Frequent bedtime stories a Rare theater a Raw gap [S.E.] [0.02] [0.04] [0.02] [0.02] [0.02] Conditional gap [S.E.] [0.02]* [0.05] [0.02] [0.03]* [0.02] Rare hiking a Bedtime stories never b Bedtime stories every day b HOME cognitive HOME emotional Raw gap [S.E.] [0.02] [0.02] [0.02] [0.05] [0.04] Conditional gap [S.E.] [0.02] [0.02] [0.02] [0.05]* [0.05] The results are presented in Table 4. The raw ethnic gap is substantial for each variable except the emotional HOME index. The conditional gap, however, is either indistinguishable from zero or substantially smaller than the raw gap. While these results cannot be interpreted as causal effects, we take them as evidence supporting the overwhelming role of education and poverty in health and parenting, as opposed to intrinsic ethnic effects. 12

13 VI. CONCLUSIONS Our results show that the test score gap between Roma and non-roma 8 th graders in Hungary is similar to the Black-White gap present in the U.S. during the 1980s. After accounting for health, parenting, school and class fixed effects and family background, the test score gap disappears in reading and decreases by 85 percent in mathematics. We also showed that the large ethnic gaps in health and parenting disappear or decrease considerably if parental education and measures of family income and poverty are included. While causality is difficult to determine in our regressions, these results are consistent with the conclusion that education and poverty play an overwhelming role in the large ethnic test score gaps in Hungary, with health, parenting and schools as the key transmission mechanisms. 13

14 REFERENCES Barany, Zoltan The East European Gypsies. Cambridge University Press. Behrman, Jere R., and Mark Rosenzweig Returns to Birthweight. The Review of Economics and Statistics, 86(2): Black, Sandra E., Paul J. Devereux, Kjell G. Salvanes From the Cradle to the Labor Market? The Effect of Birth Weight on Adult Outcomes. The Quarterly Journal of Economics, Bradley, Robert H., Robert F. Corwyn, Bettye M. Caldwell, Leanna Whiteside-Mansell, Geil A. Wasserman and Iris T. Mink Measuring the Home Environments of Children in Early Adolescence. Journal of Research on Adolescence, 10(3): Brooks-Gunn, Jeanne, and Lisa Markman The Contribution of Parenting to Ethnic and Racial Gaps in School Readiness. The Future of Children, 15(1): Case, Anne, and Christina Paxson Stature and Status: Height, Ability, and Labor Market Outcomes. Journal of Political Economy, 116(3): Case, Anne, Darren Lubotsky, and Christina Paxson Economic Status and Health in Childhood: The Origins of the Gradient. American Economic Review, 92(5): Educational Testing Service. (2010). The Black-White Achievement Gap. When Progress Stopped. Princeton (NJ). Fryer, Roland G. and Steven D. Levitt. (2006). The Black-White Test Score Gap Through Third Grade. American Law and Economic Review, 8(2), p Kertesi, Gábor, and Gábor Kézdi. Forthcoming. Roma Employment in Hungary after the Post-Communist Transition. The Economics of Transition. Kemény, István, and Béla Janky Roma Population of Hungary In Roma of Hungary, ed. István Kemény, New York: East European Monographs Atlantic Research and Publications. Magnuson, Katherine and Jane Waldfogel, eds. (2008), Steady Gains and Stalled Progress: Inequality and the Black-White Test Score Gap. New York, Russell Sage. Mott, Frank L The Utility of the HOME-SF Scale for Child Development Research in a Large National Longitudinal Survey: The National Longitudinal Survey of Youth 1979 Cohort. Parenting: Science and Practice, 4(2-3): Reichman, Nancy E Low Birth Weight and School Readiness. The Future of Children, 15(1):

15 ONLINE APPENDIX Table 1A. Detalis of the sample selection. Numbers of observations and statistics on test scores and the mothers level of education. Number of observatio ns Mean standardized test score a reading mathematic s Data from the Hungarian National Assessment of Basic Competences, grade 8 Fraction with mother's education 8 grades or less college All registered students 113,092 n.a. n.a. n.a. n.a. Students with test scores in reading 109, n.a. n.a. n.a. Students with test scores in mathematics 104,566 n.a n.a. n.a. Students with test scores both in reading and mathematics 104, n.a. n.a. Students with test scores and family 88, background data Students who agreed to participate in the Hungarian Life Course Survey 37, Data from the Hungarian Life Course Survey Sample in wave 1 b 10, Sample in wave 2 b 9, Estimation sample b 9, Notes. a Test scores are standardized by official figures on national means and standard deviations. Not all students scores are included in the national statistics, therefore the nonzero means in the total population. b All statistics (mean test scores and fractions) are weighted by sampling weights. 15

16 Summary statistics Table 2A. Roma Non-Roma Variable mean sd n mean sd n Low birth weight Body height (standardized) Subjective health fair or poor Frequent bedtime stories (parent s answer) Rare theater with parents (parent s answer) Rare hiking with parents (parent s answer) Bedtime stories never (child s answer) Bedtime stories every day (child s answer) Cognitive HOME index Emotional HOME index Lives with biological mother Lives with stepmother Lives with biological father Lives with stepfather Mother s education 0-8 th grade Mother s education vocational Mother s education secondary reference Mother s education college Father s education 0-8 th grade Father s education vocational Father s education secondary Father s education college Books: less than Books: Books: Books: reference Books: Books: Books: more Internet at home reference Mother employed Father employed Fraction of years mother was employed Fraction of years father was employed ln Household income ln Household size Non-employed adults in household Square meter per capita Rooms per capita Bathroom Poverty indicator (no money for food) Poverty indicator (no money for heating) Poverty indicator (child welfare allowance) Poverty indicator (free lunch) Poverty indicator (free schoolbooks)

17 Table 3A. Detailed results of the regressions on standardized test scores in reading (1) (2) (3) (4) (5) (6) (7) Roma [0.05] [0.05] [0.05] [0.06] [0.07]* [0.05]* [0.07] Low birth weight [0.04] [0.04] [0.05]* [0.05] [0.04] [0.05] Body height (standardized) Subjective health fair or poor Frequent bedtime stories (parent s answer) Rare theater with parents (parent s answer) Rare hiking with parents (parent s answer) Bedtime stories never (child s answer) Bedtime stories every day (child s answer) [0.01] [0.01] [0.01] [0.02]* [0.01] [0.01] [0.04] [0.04] [0.04] [0.05] [0.04] [0.05]* [0.03] [0.03] [0.04] [0.02] [0.04]* [0.03] [0.03] [0.04] [0.03] [0.04] [0.03]* [0.03] [0.04] [0.03] [0.04] [0.06] [0.06] [0.07] [0.06]* [0.07] [0.02] [0.03] [0.03] [0.02] [0.03] Cognitive HOME index [0.01] [0.02] [0.02] [0.02] [0.02] Emotional HOME index [0.01] [0.02] [0.02] [0.01] [0.02] Lives with biological mother [0.05] [0.08] Lives with stepmother [0.05] [0.07] Lives with biological father [0.04] [0.06] Lives with stepfather [0.04] [0.06] Mother s education 0-8 th grade Mother s education vocational Mother s education secondary Father s education 0-8 th grade Father s education vocational Father s education secondary [0.04] [0.06] [0.04] [0.07] [0.03] [0.04] [0.27] [0.34] [0.28] [0.34]

18 [0.45] [0.49] Books: less than [0.45] [0.49] Books: [0.05] [0.07] Books: [0.04] [0.06] Books: [0.03] [0.05] Books: [0.05] [0.08] Books: [0.04] [0.06] Internet at home [0.04] [0.06] Mother employed [0.03] [0.05] Father employed [0.04] [0.05] Fraction of years mother was employed Fraction of years father was employed 18 [0.04] [0.06] [0.05]* [0.07] ln Household income [0.03] [0.04] ln Household size [0.05]* [0.08] Non-employed adults in household Apartment size, square meters per capita [0.02] [0.03] [0.00] [0.00] Rooms per capita [0.05] [0.08] Bathroom [0.06] [0.08] Poverty indicator (no money for food) Poverty indicator (no money for heating) Poverty indicator (child welfare allowance) Poverty indicator (free lunch) Poverty indicator (free schoolbooks) [0.05] [0.06] [0.03] [0.05] [0.03] [0.04] [0.04] [0.06] [0.03] [0.04] Missing birth weight [0.15] [0.13]* [0.15] [0.21] [0.14] [0.21] Missing height [0.08] [0.08] [0.09] [0.11] [0.07] [0.11] Missing subjective health [0.10] [0.09]* [0.12] [0.14] [0.08]* [0.13] Missing bedtime stories [0.06]* [0.07] [0.08] [0.06] [0.08]

19 Missing cognitive HOME index Missing emotional HOME index [0.09] [0.11] [0.14] [0.09] [0.15] [0.07] [0.09] [0.11] [0.07] [0.12] Missing number of books [0.13] [0.18] Missing Internet [0.19] [0.21] Missing education of mother Missing education of father Missing household income Missing apartment size, square meters [0.26] [0.33] [0.45] [0.50] [0.03] [0.05] [0.09] [0.13] Missing number of rooms [0.16] [0.19] Missing bathroom [0.18] [0.27] Missing poverty indices [0.11] [0.16] Constant [0.02] [0.02]* [0.03]* [0.03] [0.04] [0.57] [0.81] School FE YES YES YES School Class FE YES YES Observations R-squared

20 20 Table 4A. Detailed results of the regressions on standardized test scores in mathematics (1) (2) (3) (4) (5) (7) (6) Roma [0.05] [0.05] [0.05] [0.05] [0.07] [0.05] [0.07]* Low birth weight [0.04] [0.04] [0.05] [0.05] [0.04] [0.05] Body height (standardized) Subjective health fair or poor Frequent bedtime stories (parent s answer) Rare theater with parents (parent s answer) Rare hiking with parents (parent s answer) Bedtime stories never (child s answer) Bedtime stories every day (child s answer) [0.01] [0.01] [0.01] [0.02]* [0.01] [0.02] [0.04] [0.04] [0.05] [0.06] [0.04] [0.06] [0.03] [0.03] [0.04] [0.03]* [0.04] [0.03] [0.03] [0.04] [0.03] [0.04] [0.03] [0.03] [0.04] [0.03] [0.04] [0.06] [0.06] [0.07] [0.06] [0.07] [0.03] [0.03] [0.04]* [0.03] [0.04] Cognitive HOME index [0.01] [0.02] [0.02] [0.02] [0.02] Emotional HOME index [0.01] [0.02] [0.02] [0.01] [0.02]* Lives with biological mother [0.06] [0.09] Lives with stepmother [0.06] [0.08]* Lives with biological father [0.05] [0.07] Lives with stepfather [0.05] [0.07] Mother s education 0-8 th grade Mother s education vocational Mother s education secondary Father s education 0-8 th grade Father s education vocational Father s education secondary [0.05]* [0.07] [0.05] [0.08] [0.03] [0.04] [0.30] [0.32] [0.31] [0.33]